Aligning biomedical ontologies using lexical methods and the UMLS: the case of disease ontologies

The process of aligning ontologies comprises two major steps: i) mapping concepts and ii) characterizing the relations between the concepts. In this paper, we present an alignment method based on a hybrid approach that reuses the UMLS knowledge base and aims at identifying patterns to characterize the relations. The proposed method consist in four steps: 1) exact matching, 2) searching for terms from one ontology that are included in terms from the other ontology, 3) identifying direct relations through the UMLS and 4) extracting syntactico-semantic patterns to infer novel alignments. This method has been applied to aligning the Human Disease ontology and the Mouse Pathology ontology resulting in 48 exact matches and 3,697 pairs of concepts for which one term is included in a term from the other ontology. 1,270 alignments are present in the UMLS. Among these, 903 are characterized by a semantic attribute. Based on these alignments, a study of the syntactic patterns has been done. Not surprisingly, the distribution of the different syntactic patterns is not sufficient to discriminate the different types of relationships found in the UMLS alignments. We have used the semantic categorization of the concepts provided by the UMLS to extract syntactico-semantic patterns. 87 novel alignments based on 6 syntactico-semantic patterns associated with isa and has associated morphology have been inferred.